Dynamic Difficulty Adjustment Using Reinforcement Learning for Adaptive Gameplay Experience in Resik

Abstract

The development of educational games holds significant potential for interactively instilling environmental conservation concepts, such as reduce, reuse, and recycle (3R). However, varying player skill levels often lead to boredom when a game is too easy, or frustration when it is too difficult. Although Reinforcement Learning (RL)-based Dynamic Difficulty Adjustment (DDA) has proven effective in balancing difficulty levels in action or MOBA genres, its application in educational waste management games remains underexplored. This study aims to develop an educational game titled "Resik" and implement a DDA mechanism using the Proximal Policy Optimization (PPO) RL algorithm, enabling the game's difficulty to adjust to player proficiency adaptively and in real-time. This research employs a Research and Development (R&D) and experimental approach. The adaptation parameters include the waste spawn rate, mission timer, sorting error rate, and NPC speed. The RL model was developed and integrated into the Unity Engine for the Android platform. Evaluation was conducted through bot simulations (15 iterations) and field trials involving 50 students from SMKS Bina Bangsa Pertiwi. The integration of the PPO RL algorithm into the game engine was successfully implemented, allowing the game to respond to player performance and dynamically adjust the difficulty. The application of RL-based DDA proved effective in maintaining the gameplay experience within the adolescent players' flow zone. These findings contribute to the development of adaptive educational games while simultaneously supporting the improvement of waste management literacy.